David Klonoff: 0:13

Welcome to Diabetes Technology Report. I'm Dr David Klonoff from Mills Peninsula Medical Center in San Mateo, California. I'm here with my co-host and we have a very special guest from very far away from California. David will introduce him.

David Kerr: 0:31

Hello everyone. It's great to be with you, david Klonoff, today, and, of course, with our very, very special guest, dr Chen Wong, who's speaking to us from Beijing. We are so fortunate because you are a guru when it comes to the use of technology for the detection of diabetic retinopathy, so I'm very excited. But what we'd like to ask our guests first of all is how did you end up where you are with an interest in technology and eye disease and diabetes? What was that journey for you, dr Wong?

Tien Wong: 1:07

Okay, thank you very much. I'm so pleased to be here. My journey really began as an ophthalmologist and I was doing my PhD in the US at Johns Hopkins, and I was also doing field work and a postdoctoral fellowship at the University of Wisconsin Medicine, and at that time diabetic retinopathy was just beginning to be recognized as a major cause of blindness in the US and elsewhere in the world. Working with Ronald and Barbara Klein from the landmark Wisconsin Epidemiological Study of Diabetic Retinopathy, I went into the field whereby looking at not just the epidemiology, but how do we prevent the disease, how do we screen for the disease and therefore lead to better treatment. So a lot of my early research was focused on that. I therefore stumbled onto what I'll say is quite an important part of managing diabetic retinopathy as well as other diabetes complications which I can get into subsequently.

Tien Wong: 2:15

But at that time we were beginning to move from film-based photography to digital photography, photography to digital photography, and therefore we you know the whole group had to be very involved in understanding what does digital photography mean? How do you standardize the technology? What does a high quality digital image offer you? And then we started working with computer scientists, and that was really before the era of AI and deep learning. Right, you know very basic pixels and understanding digital retinal photographs. So I would say it's really about 20 plus years ago almost 30 years when we delved into digital diabetic retinopathy screening and I think that's the basis of our current technology today globally.

David Kerr: 3:12

So what are you up to at the moment? Presumably everything is artificial intelligence, deep learning.

Tien Wong: 3:20

So, at the moment, we have taken, I would say, a longish journey towards using the retinal images that we now take for granted.

Tien Wong: 3:33

It's all digital, right and it's taken from smaller and more portable retinal cameras, and that's for primarily two purposes.

Tien Wong: 3:42

One is, of course, to detect early stages of diabetic retinopathy and therefore to refer them, when appropriate, to the eye specialist for them to have treatment, usually at eye centers or eye clinics.

Tien Wong: 3:58

So that's a fundamental part of diabetic retinopathy screening, part of diabetic retinopathy screening. Secondly, we're using these retinal images to look at a really even newer area in this field of using the retinal photographs digital, which can allow us to look and interrogate, for example, the blood vessels in the not just in the eye but the rest of the body, as well as the nerve fiber layers. That's seen in the eye, but the rest of the body as well as the nerve fiber layers that's seen in the eye but again representative of nervous and neural systems in the rest of the body, and this whole field of using the eye as what I would say a window or a channel to understand the systemic complications of diabetes which, as everyone knows, you know, diabetes affects the heart, the brain, the kidneys and so on and so forth. So that allows us to use a single technology, in other words a digital retinal photograph, to not only understand eye diseases but also, probably, the systemic complications of diabetes, and that's a very exciting new field.

David Klonoff: 5:10

Tian, can you comment on the idea of oculomics and how it will be used to evaluate and treat people with?

Tien Wong: 5:18

diabetes. So oculomics is again the term coined, whereby we are using the eye images and the information in the eye, so the omics from the eye or the ocular structure, and using it to interrogate the different systemic complications related to diabetes, and even in people without diabetes. But let's concentrate on diabetes. So, for example, there has been a very strong biological and historical link between what happens in the eye and what happens in the kidneys in people with diabetes. Almost concurrently, people that have diabetic retinopathy, but particularly the severe stages, would also have kidney damage, diabetic nephropathy. So using a single screening, I would say biomarker, in other words, information from the retinal images, might offer you simultaneous understanding of damages seen in the eye as well as the rest of the body. So this is the fundamental concepts of oculomics. So no different from understanding, for example, a gene, whether it affects both the eye and the heart and the kidneys. So what we call genomics or protein, what we call proteomics, and this is now what we call oculomics.

Tien Wong: 6:44

Now, the potential of oculomics is number one. The retinal photograph is digital, it is non-invasive. The technology to capture this information happens in seconds and could be, in the future, widely available in what I would say easily accessible settings, For example, in the optometrist shop. There's been discussion that it could be even in a pharmacy, it could be in the supermarket and it could be in your neighborhood post office. So what we are potentially doing is to take a single technology that could be cheaply and widely available towards what we really need to do, which is to have not just early screening but detection of the different stages of complication that diabetes affects across. You know, US as well as the world.

David Klonoff: 7:44

Ken, in the US, a problem that we've seen is that even with screening, people often don't have follow-through to see the eye doctor for the treatment. Are you working on some type of a follow-through program so it's not just here's your diagnosis, but getting people in for treatment?

Tien Wong: 8:03

And this is the new, exciting era of large language models.

Tien Wong: 8:07

As most of us in the world now are aware, the large language models offers what we call very tailored, very specific instructions that can be used to help patients understand their problem, understand the severity of their condition, as well as have very targeted recommendations for them to either have follow-up, for example, with the eye doctors or with their kidney doctors, or to have better control of their sugars and blood pressure.

Tien Wong: 8:43

And therefore, in one of our recent papers, we have combined what we call the retinal images with a large language model component, so that they extract the information from the patient, as well as the retinal images, and give a targeted, specific advice that might hopefully prompt the patient to take better control of their own health situation. Now, this is different from what was previously, prior to the era of large language model, prior to large chat, gpts and so forth, whereby people would get a pamphlet, which will be good advice, but it will be very generic advice and it doesn't really, I would say, tailor to a specific individual or patient. And we think that in the era of AI, digital imaging combined with large language models, we should be better able to tackle this part, which is that patients get information they don't know what to do with it. They're not motivated to do anything with it and hopefully this large language model offers that patient assistance.

David Kerr: 9:54

Tian, this is absolutely fascinating something that we're very interested in. My world at Sutter Health here in California, we're also contemplating whether you can combine more frequent retinal imaging with, say, other wearables such as continuous glucose monitors or blood pressure monitoring, and so you can actually have, from a perspective of time you can see, changes in the retina which reflect changes in glucose. Are you involved in work in that area as well?

Tien Wong: 10:28

I think a lot of groups are working combining multiple modalities, and certainly we have some work with people on the wearables field, as well as those using what we'll call mobile technology, and information that feeds this mobile technology will likely come from multiple places, right From the standard doctor's notes to the laboratory records, continuous glucose monitoring and, of course, you know, the periodic visits for us will be the optometrist and for some people would be the kidney specialist and I think, ultimately, an individual patient will, at the point of diagnosis, have a digital record of information that's collected over time from various sources, no different from, you know, from our own digital history.

Tien Wong: 11:41

I mean, we have our own Facebook history, our Instagram history I likely see a little bit of this our diabetes history. What we are not seeing yet, it's a holistic platform whether you can call it an app or a device that is able to extract these very diverse sources of data and to make sense of these data. But I think that will likely come soon. It's really what we call a new field, where people say it's convergent science. Right, you know, we have science that is very siloed, kind of developing on their own, and what we need is a platform that brings together data and technology, ai you know large language models into something simpler. You know we need a little bit like a Steve Jobs iPhone moment right, whereby we're able to have that single companion, that buddy that helps us with our health over time.

David Kerr: 12:51

Essentially and I think that that's something that you know, we wish to see for our patients and for our providers- so, just following on from that, on a more negative aspect of that, if you take 100 people and you try and look at the back of the eye without dilating the pupils a proportion of them you just don't get very good images. Is that a problem that's always going to be there, or do you think the technology is getting better and better, that the ungradable or unviewable images is going to be zero cameras? They?

Tien Wong: 13:27

were poor quality images and then newer cameras have tracking devices, there will likely be some sort of automatic kind of artifact removal. There will be likely some generative parts of the images that will be synthetic but will follow the pattern of what the large language models is predicting generative AI. So you don't need all the information on everything right I think you need substantial information and it will be able to produce an image that is as close as possible to the real image. So I think that technology whereby the new generative AI algorithms allow will herald an era whereby most of the images will be gradable and readable and interpretable for everyone.

David Klonoff: 14:35

Tian, one of the most fascinating areas of AI is agent hospital. Could you talk about that, what you're doing and what the hospital is all about?

Tien Wong: 14:46

We are starting to think about how do we build something that is for healthcare using AI, and there are really, I would say, three approaches, right? The first approach is the traditional hospital, the traditional clinics. All of us are very familiar with it. They are not nice places to visit. They are usually crowded, there are usually lots of gaps in the care. We spend three hours in the clinic and we see our physician for maybe five minutes, right, so there are many challenges that happens in the clinic and we see our physician for maybe five minutes, right, so there are many challenges that happens in the traditional hospital. So one way, of course, is to transform this hospital by putting in more applications, having back-end AI tools, using some front-end AI tools where, before the patients see the doctors, they have an interview, so-called with an AI chatbot, right, so those are being done. What I would say traditional hospital plus small AI. Now, is that doable? Is that going to be successful?

Tien Wong: 15:51

I think there's many, many experiments going on everywhere and we have yet to see major transformation. In fact, there's probably a lot of mental fatigue, there's probably increase in costs, there's uncertainty from patients and physicians and one app leads to another and there's very little convergence. So what we are trying to do is to say let's build it from scratch, right? Almost a hospital that is built digitally from beginning, and that's the concept of the AI agent hospital, and therefore that hospital should have, in one sense, co-development by physicians and engineers, an idea whereby the financial model or the sustainability of the model is not dependent on how many patients come in, how many procedures are done, how many tests are done, but on a model in which we are able to integrate the most cost-effective digital platforms and artificial intelligence with as few touch points by the physicians as possible.

Tien Wong: 16:59

In fact, we envision that at some stage, half, maybe even more, of these patients will not need to interact with a physical healthcare provider, right?

Tien Wong: 17:10

So you need to start from that kind of basis, and that's where we are not calling it an AI hospital, but an AI agent hospital, whereby the agent is the primary coordinator of care within the hospital and therefore, as I said, it's envisioned that 50% probably do not ever need to see that physical human healthcare provider or physician for that matter and therefore we hope to have targeted improvement in efficiency, of course, in diagnosis and safety. That's the basis of the training the agent, but also things that we don't really have a good handle of, you know, the throughput of patients, the shortness of the waiting times in the hospital and, ultimately, whether they even need to come to the hospital, because we can imagine that the AI agent hospital a lot of those care that's provided will be via mobile devices, will be via telemedicine and in their homes, right. So that requires a complete thinking. The AI agent hospital really needs a concept that we do not yet know, but we are trying to build.

David Klonoff: 18:33

Tian. One topic that we've talked about is how do you get patients to trust the AI If they're not seeing a doctor. The patient has to really find that it's trustworthy.

Tien Wong: 18:44

Yeah, I think trust is a very key component of the entire AI ecosystem. I don't think we pay sufficient attention to that. We usually pay number one, the technology development. Right, how robust? How big is the data set? That's training the algorithm. We looked at another aspect, which is the clinical outcomes. Right, the clinical outcomes. Is it efficacious? Is it doing what it's supposed to do? Is it picking up the disease? Is it missing cases? But in between is what I would say a very important area, which is on this entire system of trust, and you can say it is behavioral, it is psychological, it is experiential and it involves multiple stakeholders.

Tien Wong: 19:36

Now, how do we define trust? You can say that we traditionally have put a trust in the patient-doctor relationship. Now you are adding a third party involved in this patient-doctor, so you need to build a trust in this three-party relationship. It's not that easy, right? I mean, we know who we like, who we can interact with. Do we want someone else in the room? Right?

Tien Wong: 20:04

And I think that, therefore, that trust needs to be from a multi-faceted approach. Number one patient needs to trust. And what does the patient trust mean? The doctor needs to trust because you know there's another person in the room. So there's also a healthcare provider trust relationship, and then the system needs to trust, because the system needs to know that we're building a system whereby it's not just a two-way interaction between the patient and the healthcare professional, but another part in the system, and the system's trust could involve many different things that we are not able to see. So I think that a whole idea of trust is something that we need. Other groups of stakeholders involved in this AI relationship, and currently I see two major stakeholders the computer scientists, the engineers and the physicians who want to use the technology. We need that third group in the room.

David Kerr: 21:03

Tian. I have one final question, just to bring it back to the retinal imaging. I was brought up that the retina is the gateway to the soul. If you move beyond diabetes and the complications, what other diseases do you think are going to be detectable at the back of the eye at a very early stage?

Tien Wong: 21:24

going forward, I think that one of the most promising aspects beyond diabetes, is using the eye as really a marker of aging and brain health. I think these are the two most important things we are now interested in healthy aging. You can say it's longevity or it is better quality of life as we age, and I think the retina offers a lot of promise for that. For example, in one of the algorithms we have developed, what we call a retinal-phenol age. It's really an age of the person based on the changes or the health or the damage of the retina.

Tien Wong: 22:15

I will have to tell you that when I did that algorithm, my retinal age is five years older than my chronological age, which is not a good sign, essentially, right. So I'm kind of older than my own timeline should be by chronological age. I've seen people that have retinal age that is five or ten years younger than their chronological age. So, in other words, the chronological age it's just a time, it's a number, whereas the retinophenol age is a biological marker of our body essentially, and I think that that's a very interesting and very exciting era as we move into. As I said, the interest of many older people of our entire world is interested in how do we maintain healthy aging and longevity in this very complex world that we live in?

David Klonoff: 23:13

Ken, thank you very much for discussing AI, retinal health, oculomics and something I had not heard before, which is the health age from the retina. I'm going to read about that. Thank you for joining us. We hope to work with you in the future. And now I'm going to say goodbye from Diabetes Technology Report. We're available on Spotify and at the Apple Store and at the Diabetes Technology Society website. We look forward to catching up with you at our next Diabetes Technology Report. So for now, goodbye everybody. Goodbye.

Tien Wong: 23:51

Thank you very much.